CVPR2025
ScribbleLight: Single Image Indoor Relighting with Scribbles
Jun Myeong Choi, Annie Wang, Pieter Peers, Anand Bhattad, Roni Sengupta
摘要
Figure 1 . We introduce ScribbleLight, a generative model designed for indoor scene relighting from a single RGB image, that allows users to iteratively refine lighting effects in a photo using simple scribble annotations. For the first scribble example, the natural lighting coming from the window changes from nighttime to daytime by turning off both lamps and brightening the area near the window (left side) and casting a soft glow on the bed. In the second scribble example, only the right lamp is turned on. In the third scribble example, the natural lighting intensifies, thereby increasing the gloss on the bed's surface. For the final scribble example, adding the split in the light cast from the window causes the angle of the incoming light to change, creating a strong contrast with the warm glow of the bedside lamp.